Tuto pour le déploiement des sites avec Docker

A la fin de ce tuto, vous pourrez déployer votre site (ou service) sur les serveurs du BR ! Ceci permet de profiter d'une infrastructure gratuite, performante et robuste !

Note : ce tutoriel peut être délicat, n'hésitez pas à contacter un BRman pour obtenir de l'aide.

Comment déployer des conteneurs Docker sur l'infrastructure du BR

Overview

This tutorial will walk you through :

• How to convert your website to a Docker service
• How to build it automatically on Gitlab-CI
• How to deploy it automatically on our servers

Requirements

• Basics of git (videos here or the beginning of this awesome online book) ;
• Basics of containers (see a really good explanation here) and of docker-compose (see the getting-started here) ;
• Basics of web (see an introduction on MDN) ;
• Basics of gitlab-ci (reading the getting-started is recommended) ;

Existing architecture

A physical server called endalcher deploys and runs your container. It uses the following services to connect your service to the Internet:

• docker-compose to deploy the correct environment for your containers ;
• traefik as a reverse proxy to serve your site behind your endpoint, it is configured here ;
• nginx to serve your site to the outside world (as all the binets sites).

How to convert your website to a Docker service

The exact procedure will be dependant on your project. An example can be found here : qdj-Dockerfile

A Docker container is an isolated machine on which you can run your website. In this step, we will help you convert you app to be Docker-ready. This basically means installing everything your app needs inside a linux computer. You'll need to install Docker on your computer first windows-installer.

We assume your project is hosted on the BR's gitlab instance. Create a deployment directory at the root of your repository and create a file Dockerfile (with no extension) and a file docker-compose.yaml in it.

• Dockerfile

FROM hello-world

• docker-compose.yaml

# this is a docker-compose meant for local testing of the deployment image.
# for the real deployment, go to the docker-services BR gitlab project

version: "3"
services:
  app:
    build:
      context: ..
      dockerfile: deployment/Dockerfile
#    ports:
#      - "8000:8000"

Then open a terminal in this directory and type docker compose up --build. If everything went well, you see a nice message from Docker telling you everything is working !

Now, you need to make your app work in the docker container. For this, you need:

• To update the Dockerfile to install everything you need. The following works for a basic (Django) python app :

FROM python:3

# copies all the code inside the container
COPY . /my_super_project

WORKDIR /my_super_project
RUN pip install --no-cache-dir -r requirements.txt
CMD python3 manage.py runserver

• To update the docker-compose file to expose the port you need (uncomment the corresponding line in the dockerfile).

How to handle persistant data

If you app uses an external database or need to not lose some files accross resets, you should use docker volumes with or without a postgres image. The jtx-docker-compose might help you understand what you need.

How to handle secrets

Often your app has some secrets in a file (such as .env) that is not in Git. DO NOT put it in git (or all your secrets will be stolen !). Later we are going to see how to securely make them available for the automatic gitlab build.

Automatic build

We will you gitlab continuous integration to build your docker image directly on the BR's servers. Write the following into a file name .gitlab-ci.yml at the root of your repository :

# Build docker image and upload to registry
variables:
  IMAGE_TAG: $CI_REGISTRY_IMAGE:$CI_PIPELINE_IID
  PROJECT: mySuperProject

docker:build:
  stage: build
  only: # put here the git branches you want the job to run on :
    - main
    - stable
  image: docker:stable
  services:
    - docker:dind
  variables:
    DOCKER_HOST: tcp://docker:2375/
    DOCKER_DRIVER: overlay2
    DOCKER_TLS_CERTDIR: '' # disables TLS in docker-in-docker (not needed)
  script:
    - docker login -u $CI_REGISTRY_USER -p $CI_REGISTRY_PASSWORD $CI_REGISTRY
    - docker build --pull --build-arg version=$CI_PIPELINE_IID -t $IMAGE_TAG -f deployment/Dockerfile .
    - docker push $IMAGE_TAG

As you commit this file to Gitlab, you should see a new job appear (or look under Build/jobs). Your image should build correctly (if not, it means something is not available to gitlab that is when you build on your computer. Read the job logs and try to fix it !)

Checking the image

This step is optional, but it is the easiest way to debug your image ! You can run it locally like so :

docker login registry.binets.fr
docker run --rm registry.binets.fr/<nom.prenom>/<my-gitlab-repo>:<image>

Where image is the number of the image (available in Gitlab -> Deploy -> Container registry) and <nom.prenom>/<my-gitlab-repo> is the same as in your project Gitlab URL.

Typically, your app won't start because you forgot a secret file (see below how to manage that !).

Automatic deployment

We will now add a second CI job to your repository to deploy your image. It will trigger another CI pipeline of br/docker-services. This pipeline has the rights to deploy on endalcher, and do so using a docker-compose.yaml file. To trigger it, add the following to your .gitlab-ci.yml

docker:deploy:
  stage: deploy
  only:
    - stable
  image: debian:latest
  when: manual
  script:
    - apt-get update && apt-get install -y curl
    - curl --fail-with-body --request POST --form token=$TRIGGER_TOKEN --form ref=master --form variables[PROJECT]=$PROJECT --form variables[DOCKER_IMAGE]=$IMAGE_TAG https://gitlab.binets.fr/api/v4/projects/560/trigger/pipeline

Make sure the variable PROJECT in .gitlab-ci.yml is correct : our pipeline will deploy the file named docker-compose.${PROJECT}.yaml.

In order to create the correct docker-compose.${PROJECT}.yaml file, you will need to create a Merge Request on br/docker-services, or send a docker-compose.yaml file to a BRman. The br/docker-services is full of examples and a basic one is explained below :

Extra parameters and secrets

You can also pass any parameter needed in the compose file by adding arguments to the cURL call like so : --form variables[MY_SUPER_VARIABLE]="elle vaut 42!"

If you want to pass secret environnements variables, store then in gitlab under Settings/CI-CD/Variables and make sure they are "protected" and "masked". Pass them like so --form variables[MY_PASSWORD]=$PASSWORD_STORED_IN_GITLAB.

Be aware that any secret passed this way is visible to the BR and everyone that has maintainer access to you repository ! If you need help setting up, don't hesitate to ask for help from the BR !

Secret files

Be aware that any secret passed this way too is visible to the BR and everyone that has ANY access to you repository (because they can get the docker image) ! So only do this for a private registry or make sure your setting Settings -> General -> Visibility, project features, permissions only allows for registry access for members

Another way to store secrets in your Gitlab repo is Gitlab -> Settings -> CI-CD -> Secure-files. You can upload any (small) secret file here and it will not be exposed ! You can downlod it at buildtime like so :

# Build docker image and upload to registry
docker:build:
  stage: build
  only:
    - main
    - stable
  image: docker:stable
  services:
    - docker:dind
  variables:
    DOCKER_HOST: tcp://docker:2375/
    DOCKER_DRIVER: overlay2
    # For disabling TLS in docker-in-docker (not needed)
    DOCKER_TLS_CERTDIR: ""
    # !!! where to download our secure .env file 
    SECURE_FILES_DOWNLOAD_PATH: "."
  script:
    # !!! download secure .env file
    - apk add --no-cache curl bash
    - curl --silent "https://gitlab.com/gitlab-org/incubation-engineering/mobile-devops/download-secure-files/-/raw/main/installer" | bash
    # build docker image
    - docker login -u $CI_REGISTRY_USER -p $CI_REGISTRY_PASSWORD $CI_REGISTRY
    - docker build --pull --build-arg version=$CI_PIPELINE_IID -t $IMAGE_TAG -f deployment/Dockerfile .
    - docker push $IMAGE_TAG
    - echo "=> pushed $IMAGE_TAG"

Private repository

Keeping your repository private can be beneficial for security or privacy. But it require an extra step for the BR to access the docker image. We wille pass a CI deploy token as cURL parameter.

• Go to your Settings -> Repository -> Deploy token and create a token name exactly gitlab-deploy-toke

• Add the following parameters to the cURL call in your CI deploy job : --form variables[OPTIONAL_PRIVATE_CI_REGISTRY]=$CI_REGISTRY --form variables[OPTIONAL_PRIVATE_CI_REGISTRY_USER]=$CI_DEPLOY_USER --form variables[OPTIONAL_PRIVATE_CI_REGISTRY_PASSWORD]=$CI_DEPLOY_PASSWORD

Example

You want to deploy the QDJ site with docker, and bind it to the server name qdj.binets.fr. Your site is running on flask and uses a postgresql database.

Pre-existant configuration

You have a little Dockerfile:

# Base image: python3
FROM python:3
WORKDIR /usr/src/app

# Install dependacies
COPY requirements.txt ./
RUN pip install --no-cache-dir -r requirements.txt

# Copy application files
COPY . .

EXPOSE 5000/tcp

# the command runs the migrations and then launches gunicorn to serve the app
CMD python -m flask db upgrade && gunicorn -b :5000 qdj:app

You have the following docker-compose.yaml:

version: '3.1'

services:
  db:
    # Use a small postgres image, as we don't need anything on this container
    image: postgres:alpine
    environment:
      POSTGRES_USER: qdj
      POSTGRES_PASSWORD: somepassword
      POSTGRES_DB: qdj
    # store the data on a persistent file
    volumes:
      - "qdj-data:/var/lib/postgresql/data"
    restart: always

  app:
    # Use the image from Dockerfile
    image: ${DOCKER_IMAGE}
    environment:
      FLASK_APP: qdj.py
      FLASK_ENV: production
      FLASK_DEBUG: 0
      DATABASE_URI: "postgresql://qdj:somepassword@db:5432/qdj"
      SECRET_KEY:
    # Tells to docker to start app after db (otherwise flask crashes)
    depends_on:
      - db
    # Always restart on crash
    restart: always
    
volumes:
  qdj-data:

Production docker-compose.yaml

We will adapt your docker-compose.yaml file in order to bind your site to the hostname :

1. First let's connect your app container to the global network. We won't bind the db container to the global network to protect it from intruders. Network graph in endalcher We will add network configuration to the docker-compose.yaml:

   • to enable the network web, we add at the bottom of the file:

     networks:
       web:
         external: true
     

   • to connect app to the network web, we add to the app configuration:

     networks:
       - web
       - default
     

2. Then we can add labels to tell traefik to match the hostname to a packet going to app:

   labels:
       - "traefik.docker.network=web"
       - "traefik.frontend.rule=Host:qdj.binets.fr"
       - "traefik.port=5000"
   

This gives us the following docker-compose.qdj.yaml:

version: '3.1'

services:

  db:
    image: postgres:alpine
    restart: always
    environment:
      POSTGRES_USER: qdj
      POSTGRES_PASSWORD: qdjpw
      POSTGRES_DB: qdj
    volumes:
      - "qdj-data:/var/lib/postgresql/data"

  app:
    image: ${DOCKER_IMAGE}
    environment:
      FLASK_APP: qdj.py
      FLASK_ENV: production
      FLASK_DEBUG: 0
      DATABASE_URI: "postgresql://qdj:qdjpw@db:5432/qdj"
      SECRET_KEY:
    depends_on:
      - db
    restart: always
    networks:
      - web
      - default
    labels:
      - "traefik.docker.network=web"
      - "traefik.frontend.rule=Host:qdj.binets.fr"
      - "traefik.port=5000"

volumes:
  qdj-data:

networks:
  web:
    external: true

This file goes into the br/docker-services repository, it will be used by the triggered pipeline to create your containers.

Triggering pipelines with Gitlab CI

In this pipeline, we will need to add 2 jobs:

• one to build the docker image using the Dockerfile
• one to trigger the br/docker-services pipeline to create the containers.

To build the docker image, we will use the template do so:

# Build docker image and upload to registry
docker:build:
  stage: build
  only:
    - master
    - stable
  image: docker:stable
  services:
    - docker:dind
  variables:
    DOCKER_HOST: tcp://docker:2375/
    DOCKER_DRIVER: overlay2
    # For disabling TLS in docker-in-docker (not needed)
    DOCKER_TLS_CERTDIR: ''
  script:
    - docker login -u $CI_REGISTRY_USER -p $CI_REGISTRY_PASSWORD $CI_REGISTRY
    - docker build --pull --build-arg version=$CI_PIPELINE_IID -t $IMAGE_TAG -f deployment/Dockerfile .
    - docker push $IMAGE_TAG

The second job is rather tricky. We will use curl to trigger the pipeline, using Gitlab Pipeline API. This call is secured by a trigger token, which should be saved into your project variables as $TRIGGER_TOKEN. This is what the job looks like:

docker:deploy:
  stage: deploy
  when: manual
  script:
    - apt-get update && apt-get install -y curl
    - curl --request POST --form token=$TRIGGER_TOKEN --form ref=master --form variables[PROJECT]=qdj --form variables[DOCKER_IMAGE]=$IMAGE_TAG https://gitlab.binets.fr/api/v4/projects/560/trigger/pipeline

Pay attention to this syntax because debugging this api call takes a lot of time...

If you want to add variables that will be used in your docker-compose.yaml, add other arguments to curl like --form variables[WELCOME_MESSAGE]="Hello World!".

You can see an example of a CI using this in the br/qdj repo.

Note: for BR project, you can use the trigger syntax in the API call since you don't need autorisation.

History

• In 2017, the BR16 set up a Kubernetes cluster. They managed to set up a workflow to deploy seamlessly from gitlab on that cluster. However, the configuration was hard to maintain and the BR17 was not able to keep it going.
• In 2018, the BR17 (Oliver Facklam + Hadrien Renaud) set up a small docker service on a computer from the Salle Informatiques

Developping a site for BR docker deployement

Docker allows for most technologies to be abstracted into a container and thus most - if not all - sites can be deployed in docker. However, here is a bit of advice to make deployement less painful :

• Before using a particular external service (something that should run in a seperate container), check that a well supported docker image exists and with a BRman that there are no other constraints (for example, using a MinIO will go through a seperate dedicated server that already exists rather than a container).
• Use a postgres database to get access to an admin website for your server.